Matthew Byng-Maddick wrote:
> 
> On Tue, Jun 19, 2001 at 11:41:45AM +0100, Dominic Mitchell wrote:
> > On Tue, Jun 19, 2001 at 11:39:32AM +0100, Leon Brocard wrote:
> > > Greg Cope sent the following bits through the ether:
> > > > I want to design a mailer for sending large numbers of individual
> > > > messages to a large list.
> > > You're writing a mailer in Perl. Mailers have been done before. If
> > > you're using a slow one, then look at other ones, such as exim.
> 
> What you actually want to do is slightly different, you want to send
> your mail onto the queue in a way such that exim will deal with it in
> the most sensible way.
> 
> This would probably involve sorting by reverse domain, as queuing strategies
> with MX lookups can get rather complicated, and then queueing about 100
> recipients at a time. The number 100 is pulled out of thin air.
> 
> > The trouble is that practically all mailers will write the queue file to
> > disk before they do *anything* else, for safety reasons.  If I
> > understood the initial message correctly, this is his problem.
> 
> Ah, but if the mails are all the same, then there is only one queue file
> per group of recipients...
> 

No all messages are different (A Different Dear Jo line, etc ...)- if
they were the same I'd use ezmlm which flies for large lists.


> > I think that qmail, exim and postfix all fall into this category.  I
> > have a feeling that it can be turned off in sendmail, but you'd need to
> > check.
> 
> Not sure about this. qmail is a *very* bad idea if you're doing huge mail
> shots. It decides to ignore the SHOULD in RFC1123 which says you should
> do
> | RCPT TO
> | RCPT TO
> :
> | DATA
> 
> and instead opens up as many connections as it can to the remote mailserver
> wasting bandwidth and hammering both ends.
> 

DJB claims that its just as fast - and the above assumes all message
bodies are the same.

> Its delivery strategy after a backlog is also crap, because it's just FIFO.
> 

I've not had a problem with this, but my hosts are setup a little
special (low timeouts for SMTP responses and queue lifetime)

> MBM (really really anti-qm**l)

I'm sure you'll see the light one day :-)

Thanks

Greg
> 
> --
> Matthew Byng-Maddick         <[EMAIL PROTECTED]>           http://colondot.net/

Reply via email to